CN115049171B - Photovoltaic power prediction method and system based on feature migration - Google Patents
Photovoltaic power prediction method and system based on feature migration Download PDFInfo
- Publication number
- CN115049171B CN115049171B CN202210983674.2A CN202210983674A CN115049171B CN 115049171 B CN115049171 B CN 115049171B CN 202210983674 A CN202210983674 A CN 202210983674A CN 115049171 B CN115049171 B CN 115049171B
- Authority
- CN
- China
- Prior art keywords
- data
- photovoltaic power
- matrix
- prediction
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 38
- 238000013508 migration Methods 0.000 title claims abstract description 27
- 230000005012 migration Effects 0.000 title claims abstract description 27
- 238000012549 training Methods 0.000 claims abstract description 57
- 238000010248 power generation Methods 0.000 claims abstract description 25
- 230000002159 abnormal effect Effects 0.000 claims abstract description 11
- 238000004458 analytical method Methods 0.000 claims abstract description 11
- 239000011159 matrix material Substances 0.000 claims description 53
- 238000013507 mapping Methods 0.000 claims description 11
- 238000000354 decomposition reaction Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 230000009466 transformation Effects 0.000 claims description 5
- 238000012950 reanalysis Methods 0.000 claims description 4
- 239000000126 substance Substances 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 7
- 230000005855 radiation Effects 0.000 description 6
- 238000012706 support-vector machine Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 238000005065 mining Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/004—Generation forecast, e.g. methods or systems for forecasting future energy generation
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
- H02J3/38—Arrangements for parallely feeding a single network by two or more generators, converters or transformers
- H02J3/381—Dispersed generators
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2203/00—Indexing scheme relating to details of circuit arrangements for AC mains or AC distribution networks
- H02J2203/20—Simulating, e g planning, reliability check, modelling or computer assisted design [CAD]
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J2300/00—Systems for supplying or distributing electric power characterised by decentralized, dispersed, or local generation
- H02J2300/20—The dispersed energy generation being of renewable origin
- H02J2300/22—The renewable source being solar energy
- H02J2300/24—The renewable source being solar energy of photovoltaic origin
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Economics (AREA)
- Health & Medical Sciences (AREA)
- Strategic Management (AREA)
- Human Resources & Organizations (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Marketing (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Power Engineering (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Business, Economics & Management (AREA)
- Artificial Intelligence (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Primary Health Care (AREA)
- Quality & Reliability (AREA)
- Entrepreneurship & Innovation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Development Economics (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Medical Informatics (AREA)
- Operations Research (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Supply And Distribution Of Alternating Current (AREA)
Abstract
The invention provides a photovoltaic power prediction method and system based on feature migration, which are used for acquiring historical power generation data of a photovoltaic power station, removing abnormal values and using the abnormal values as target values of model training; obtaining a training data set through source domain training sample data and target domain training sample data based on a migration component analysis method; respectively training the LSTM and the Lasso models to predict the photovoltaic power, and performing photovoltaic power combined prediction on the prediction result by using a fixed weight coefficient to obtain a final prediction result. The method improves the model prediction precision and improves the accuracy and stability of the photovoltaic power generation short-term power prediction.
Description
Technical Field
The invention belongs to the technical field of electric power prediction, and particularly relates to a photovoltaic power prediction method and system based on feature migration.
Background
The photovoltaic output fluctuates greatly along with weather changes, which brings great challenges to the operation and maintenance of the power system. With the rapid increase of the photovoltaic grid-connected scale, the prediction precision of photovoltaic power generation needs to be improved urgently in order to ensure the safe and stable operation of an energy power system. At present, scholars at home and abroad have proposed a plurality of methods such as a support vector machine, a neural network, dimension reduction clustering and the like aiming at the short-term prediction problem of photovoltaic power generation, the main idea of research is to convert the photovoltaic power generation prediction problem into a regression problem of meteorological elements and photovoltaic power generation, and based on Numerical Weather Prediction (NWP), a mapping model between the meteorological elements and the photovoltaic power generation is established, wherein the mapping model comprises a linear model and a nonlinear model, and the common method comprises the following steps: support Vector Machines (SVM), BP Neural Networks (BPneural Networks), radial basis function Neural Networks (RBF Neural Networks), and the like.
The photovoltaic power prediction model based on the traditional method has excessive characteristics, which tend to affect the accuracy of the model, and therefore, the selection of the relevant influence data of the photovoltaic power is particularly important for effectively extracting. In addition, the traditional method only considers the mapping relation between numerical weather forecast data and photovoltaic output data, the non-stationarity of space-time data is neglected through pure time analysis, and the classical machine learning method such as a neural network and a support vector machine is easy to fall into a local optimal and overfitting state in the model training process, so that the traditional prediction method is difficult to further improve the prediction accuracy of photovoltaic power.
Disclosure of Invention
The invention provides a photovoltaic power prediction method and system based on feature migration, which improve the model prediction precision and improve the accuracy and stability of photovoltaic power generation short-term power prediction.
In order to achieve the purpose, the technical scheme of the invention is realized as follows:
a photovoltaic power prediction method based on feature migration comprises the following steps:
s1, acquiring historical power generation data of a photovoltaic power station, removing abnormal values, and using the abnormal values as target values of model training;
s2, obtaining historical prediction meteorological elements corresponding to historical power generation data of the photovoltaic power station from the numerical weather forecast, extracting time and geography related characteristics of the meteorological elements, and constructing target domain training sample data;
s3, obtaining historical meteorological elements corresponding to historical power generation data of the photovoltaic power station from the ERA5 atmosphere reanalysis meteorological data, extracting characteristics related to time and geography from the meteorological elements, and constructing source domain training sample data;
s4, obtaining a training data set through source domain training sample data and target domain training sample data based on a migration component analysis method;
s5, inputting the target value and the training data set into LSTM and Lasso models respectively, and training the models after carrying out gridding search and parameter adjustment to obtain the well-trained LSTM and Lasso models;
s6, respectively predicting the photovoltaic power through the trained LSTM model and the trained Lasso model, and performing photovoltaic power combination prediction on the prediction result by using a fixed weight coefficient to obtain a final prediction result.
Further, the method for extracting time and geographic features from meteorological elements in steps S2 and S3 comprises:
s101, extracting meteorological elements;
s102, extracting characteristics related to time, converting the timestamp information of the meteorological element data into discrete time information, and adding labels of months, weeks, days and hours to each meteorological element data;
s103, extracting features related to geography, and adding inclination angle and longitude and latitude information of the photovoltaic square matrix, and labels of sine of the solar altitude angle and cosine of the solar incident angle to each meteorological element data.
Further, the specific method of step S4 includes:
s201, constructing a characteristic number k 1 Target field training sample data ofCorresponding label;
Wherein x is Ti Is the ith item of target domain sample data; y is Ti Is x Ti A corresponding label; i =1,2 t ,;n t The number of samples in the target domain is;
Wherein x Sj Is the jth source domain sample data; y is Sj Is x Sj A corresponding label; j =1,2 s ;n s The number of the source domain samples is;
according to X S And X T Respectively constructing a kernel matrix K and a matrix L, and constructing a central matrix H;
wherein, K S,S ,K T,T ,K S,T ,K T,S Kernel functions of source domain data, target domain data and mixed domain data respectively;
wherein the content of the first and second substances,is a matrix of the units,is a full l column vector;
r is all-real-field (n) s +n t )*(n s +n t ) An order matrix; l T A transposed matrix of l;
s202, matrix pair (KLK + mu I) -1 KHK carries out characteristic decomposition, and combines the eigenvectors corresponding to m maximum eigenvalues together to obtain a projection matrix W, wherein m<(k1,k2) min ,μ>0 is a trade-off parameter;
s203, enabling X to pass through W S And X T Projective transformation into a data set of a mapping spaceAndwhile maintaining the main characteristic attributes in the data sets of the source domain and the target domain, reducing the edge distribution difference between the domains by reducing the sample mean distance between the domains;
Wherein W T Is a transposed matrix of W.
Further, in step S6, the setting method of the fixed weight coefficient includes:
first, the absolute error of each single model is calculated, and the formula is as follows:
(ii) a Where Yi is the actual power at time i, P it The predicted power of the t-th single model at the moment i is obtained, and N is the total number of samples for solving the fixed weight coefficient;
then, according to the formulaCalculating the weight w of each single model t The number of the single models is T, and T is ≧ 2T. It can be seen that the larger the absolute error of the single model is, the smaller the combination weight thereof is, and conversely, the larger the weight is.
The invention also provides a photovoltaic power prediction system based on feature migration, which comprises:
the target value module is used for acquiring historical power generation data of the photovoltaic power station, removing abnormal values and taking the abnormal values as target values of model training;
the target domain module is used for acquiring historical prediction meteorological elements corresponding to historical power generation data of the photovoltaic power station from the numerical weather forecast, extracting characteristics related to time and geography from the meteorological elements and constructing target domain training sample data;
the source domain module is used for acquiring historical meteorological elements corresponding to historical power generation data of the photovoltaic power station from the ERA5 atmospheric re-analysis meteorological data, extracting characteristics related to time and geography from the meteorological elements and constructing source domain training sample data;
the TCA module is used for obtaining a training data set through source domain training sample data and target domain training sample data based on a migration component analysis method;
the training module is used for inputting the target value and the training data set into LSTM and Lasso models to be trained respectively, and training the models after gridding search and parameter adjustment to obtain the well-trained LSTM and Lasso models;
and the prediction module is used for respectively predicting the photovoltaic power through the trained LSTM model and the trained Lasso model, and performing photovoltaic power combined prediction on the prediction result by using a fixed weight coefficient to obtain a final prediction result.
Further, a feature extraction unit is arranged in each of the target domain module and the source domain module, and the feature extraction unit includes:
the extraction subunit is used for extracting meteorological elements;
the time characteristic subunit is used for extracting characteristics related to time, converting the timestamp information of the meteorological element data into discrete time information, and adding labels of months, weeks, days and hours to each meteorological element data;
and the geographic characteristic subunit is used for extracting geographic related characteristics and adding the inclination angle and longitude and latitude information of the photovoltaic square matrix, and the labels of the sine of the solar altitude angle and the cosine of the solar incident angle to each meteorological element data.
Further, the TCA module includes:
matrix unit: constructing a feature number of k 1 Target field training sample data ofCorresponding label;
Wherein x Ti Is the ith sample data of the target domain; y is Ti Is x Ti A corresponding label; i =1,2 t ,;n t The number of samples in the target domain is;
Wherein x Sj Is the jth source domain sample data; y is Sj Is x Sj A corresponding label; j =1,2 s ;n s The number of the source domain samples is;
according to X S And X T Respectively constructing a kernel matrix K and a matrix L, and constructing a central matrix H;
wherein, K S,S ,K T,T ,K S,T ,K T,S Are respectively provided withKernel functions of source domain, target domain and mixed domain data;
wherein, the first and the second end of the pipe are connected with each other,is a matrix of the units,is a full l column vector;
r is all-real-field (n) s +n t )*(n s +n t ) An order matrix; l T A transposed matrix of l;
a characteristic decomposition unit: pair matrix (KLK + mu I) -1 KHK carries out characteristic decomposition, and combines the eigenvectors corresponding to m maximum eigenvalues together to obtain a projection matrix W, wherein m<(k1,k2) min ,μ>0 is a trade-off parameter;
a data set unit: by W so that X S And X T Projective transformation into a data set of a mapping spaceAndwhile maintaining the main characteristic attributes in the data sets of the source domain and the target domain, reducing the edge distribution difference between the domains by reducing the sample mean distance between the domains;
Wherein W T Is a transposed matrix of W.
Further, the prediction module includes a fixed weight coefficient unit, and the fixed weight coefficient unit includes:
calculating the absolute error of each single model, wherein the formula is as follows:
(ii) a Where Yi is the actual power at time i, P it The predicted power of the t-th single model at the moment i is obtained, and N is the total number of samples for solving the fixed weight coefficient;
according to the formulaCalculating the weight w of each single model t The number of the single models is T, and T is ≧ 2T. It can be seen that the larger the absolute error of a single model is, the smaller the combination weight thereof is, and conversely, the larger the weight is.
Compared with the prior art, the invention has the following beneficial effects:
(1) The invention uses the TCA method to minimize the maximum mean difference of the ERA5 sample and the numerical weather prediction sample in the feature subspace, obtains a dimension reduction feature subspace, and reduces the distribution difference among the fields in the space, thereby realizing the cross-field migration of the ERA5 data and effectively improving the model prediction precision;
(2) According to the invention, the combined prediction is carried out on the prediction results of the Lasso linear regression model and the nonlinear LSTM neural network, so that the accuracy and robustness of model prediction are improved.
Drawings
FIG. 1 is a schematic flow diagram of an embodiment of the present invention;
FIG. 2 is a predicted power versus power graph for an embodiment of the present invention.
Detailed Description
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
The design idea of the invention is as follows:
1. realizing characteristic migration of ERA5 reanalysis data and numerical weather forecast based on TCA migration component analysis:
the ERA5 reanalysis data is meteorological data obtained after quality control of atmospheric observation data (including ground observation, satellites, radar, sounding and the like) and assimilation into a global mode, has the characteristics of long time sequence, wide spatial distribution and high accuracy, has high data application value, and introduces a TCA characteristic migration method for mining deeper mapping relation between the meteorological data and photovoltaic power, so that valuable migration components are effectively extracted from data belonging to different but related fields, and a mapped source domain labeled sample is trained to improve model prediction precision.
Since ERA5 data cannot be predicted in the future, the accuracy of the ERA5 data is much higher than that of a forecast field (a result of numerical weather forecast), so that data distribution between two data sets is different but very similar, and the data can be migrated by adopting a method based on feature representation. The TCA (migration component analysis) method is that appropriate common description is searched between a source domain and a target domain, and the optimal space mapping relation is obtained, so that the distribution difference between the domains is reduced in a mapped feature subspace, but the original main feature attributes of data are not damaged, the migration of the source domain knowledge to the target domain is realized, and the classification precision of target tasks is improved.
2. Using combined prediction:
the long-term memory network (LSTM) can not only mine the spatial and time sequence relevance between output and relevant input variables, has stronger fitting capability to a big data-driven model training process, is not easy to fall into local optimum and overfitting, and is more suitable for mining the internal relevance and potential rules of weather state change and photovoltaic power fluctuation. According to the method, the prediction results of the Lasso linear regression model and the nonlinear LSTM neural network are combined for prediction, so that a satisfactory result can be obtained compared with the single model prediction, the prediction risk of the single model is dispersed, the advantages of the linear and nonlinear prediction models are combined, the generalization capability of the prediction model is further improved, and the accuracy and stability of the photovoltaic power generation short-term power prediction are improved.
Based on the above design concept, the photovoltaic power prediction method based on feature migration, which is provided by the invention and shown in fig. 1, includes:
1. acquiring the generated output data of the photovoltaic power station, removing abnormal values and using the data as a target value of model training,y i Is the target sample, n is the total number of samples, i =1,2, ·, n;
2. because photovoltaic power generation is related to time, local longitude and latitude, solar irradiance on a square matrix surface, cloud amount, wind speed and wind direction, the extraction and characterization mode of key features is very important for finding out the periodic rule among data and improving the convergence speed of model training, and the features are extracted according to the following modes:
(1) Extracting meteorological predicted elements such as: solar horizontal plane total radiation, inclined plane total radiation, solar direct radiation, solar scattered radiation, low-medium-high cloud cover, ground layer and high-layer wind speed, wind direction (the wind direction is characterized by wind direction cosine and wind direction sine), temperature, humidity, pressure and the like;
(2) Extracting characteristics related to time and geography, converting timestamp information of meteorological data into discrete time information through processing, and adding labels of months, weeks, days and hours to each piece of data;
(3) And extracting the information of the inclination angle and the longitude and latitude of the photovoltaic square matrix, and adding feature labels of the sine of the solar altitude angle and the cosine of the solar incident angle.
3. Obtaining numerical weather forecast data, said numerical weather forecast data and method of 2The photovoltaic power station power generation output data correspond to each other; constructing a feature number of k 1 Target field training sample data ofCorresponding label(ii) a Wherein x Ti The item i is labeled domain sample data, which contains various characteristics, such as total radiation, straight radiation, wind speed, wind direction, temperature and the like, and also contains some characteristic quantities of the structure; y is Ti Is x Ti A corresponding label; i =1,2 t ,;n t The number of samples in the target domain is; sample data as a test set;
4. acquiring ERA5 atmospheric re-analysis meteorological data of a target area where the photovoltaic power station is located, extracting corresponding meteorological elements of historical long-time sequences at the site of the photovoltaic power station according to the method in 2, and constructing a characteristic number k 2 Source domain sample data of,k2>k1, corresponding tag;
Wherein x Sj Is the jth source domain sample data; y is Sj Is x Sj A corresponding label; j =1,2 s ;n s The number of the source domain samples is;
5. according to X S And X T Respectively constructing a kernel matrix K and a matrix L, and constructing a central matrix H;
wherein, K S,S ,K T,T ,K S,T ,K T,S Kernel functions of source domain data, target domain data and mixed domain data respectively;
wherein, the first and the second end of the pipe are connected with each other,is a matrix of the units,is a full l column vector;
r is all-real-field (n) s +n t )*(n s +n t ) An order matrix; l T A transposed matrix of l;
6. pair matrix (KLK + mu I) -1 KHK carries out characteristic decomposition, and combines the eigenvectors corresponding to m maximum eigenvalues together to obtain a projection matrix W, wherein m<(k1,k2) min ,μ>0 is a trade-off parameter.
7. By W so that X S And X T Projective transformation into a data set of a mapping spaceAndwhile maintaining the main characteristic attributes in the data sets of the source domain and the target domain, reducing the edge distribution difference between the domains by reducing the sample mean distance between the domains;
Wherein W T Is a transposed matrix of W.
8. Will train the data setInputting and respectively training an LSTM model and a Lasso model, and training the models after carrying out gridding search and parameter adjustment to obtain the trained LSTM model and the trained Lasso model;
9. projecting a target domain test set into a mapping spaceAnd substituting the prediction result into the trained LSTM model and the Lasso model for prediction respectively, and performing photovoltaic power combined prediction on the prediction result by taking 0.7 and 0.3 as fixed weight coefficients to serve as a final prediction result.
The setting method of the fixed weight coefficient comprises the following steps:
first, the absolute error of each single model is calculated, and the formula is as follows:
(ii) a Where Yi is the actual power at time i, P it The predicted power of the t-th single model at the moment i is obtained, and N is the total number of samples for solving the fixed weight coefficient;
then, according to the formulaCalculating the weight w of each single model t The number of the single models is T, and T is ≧ 2T. It can be seen that the larger the absolute error of the single model is, the smaller the combination weight thereof is, and conversely, the larger the weight is.
The specific application case is as follows:
acquiring photovoltaic power generation data of a certain centralized photovoltaic power generation field with the installed capacity of 60MW, acquiring numerical prediction results, and predicting by respectively adopting a traditional prediction method and the method provided by the invention; then, respectively calculating the prediction accuracy of each method by adopting a power prediction accuracy calculation formula, wherein the power prediction accuracy calculation formula is as follows:
wherein, P Mi Is the actual generated power at time i, pp i And c, predicting the generated power at the ith moment, wherein Ci is the starting capacity at the ith moment, n is the actual generated power collection times of every 15 minutes in the daytime period, and i =1,2.
The results of the comparison of the prediction accuracy are shown in the following table:
therefore, the method provided by the invention can improve the prediction precision by nearly 1%.
As shown in fig. 2, which is a graph of the actual power, the predicted power by the conventional method, and the predicted power by the method of the present invention, it can be seen that the predicted power curve by the method of the present invention is closer to the actual power curve than the predicted power curve by the conventional method.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.
Claims (8)
1. A photovoltaic power prediction method based on feature migration is characterized by comprising the following steps:
s1, acquiring historical power generation data of a photovoltaic power station, removing abnormal values, and using the abnormal values as target values of model training;
s2, obtaining historical prediction meteorological elements corresponding to historical power generation data of the photovoltaic power station from the numerical weather forecast, extracting time and geography related characteristics of the meteorological elements, and constructing target domain training sample data;
s3, obtaining historical meteorological elements corresponding to historical power generation data of the photovoltaic power station from the ERA5 atmosphere reanalysis meteorological data, extracting characteristics related to time and geography from the meteorological elements, and constructing source domain training sample data;
s4, obtaining a training data set through source domain training sample data and target domain training sample data based on a migration component analysis method;
s5, inputting the target value and the training data set into LSTM and Lasso models respectively, and training the models after carrying out gridding search and parameter adjustment to obtain the well-trained LSTM and Lasso models;
s6, respectively predicting the photovoltaic power through the trained LSTM model and the trained Lasso model, and performing photovoltaic power combination prediction on the prediction result by using a fixed weight coefficient to obtain a final prediction result.
2. The method for predicting photovoltaic power based on feature migration according to claim 1, wherein the step S2 and the step S3 of extracting time and geography-related features from meteorological elements comprise:
s101, extracting meteorological elements;
s102, extracting characteristics related to time, converting the timestamp information of the meteorological element data into discrete time information, and adding labels of months, weeks, days and hours to each meteorological element data;
s103, extracting features related to geography, and adding inclination angle and longitude and latitude information of the photovoltaic square matrix, and labels of sine of the solar altitude angle and cosine of the solar incident angle to each meteorological element data.
3. The photovoltaic power prediction method based on feature migration according to claim 1, wherein the specific method of step S4 includes:
s201, constructing a characteristic number k 1 Target field training sample data ofCorresponding label;
Wherein x Ti Is the ith sample data of the target domain; y is Ti Is x Ti A corresponding label; i =1,2 t ,;n t The number of samples in the target domain is;
constructing a feature number of k 2 Source domain training sample data of,k2>k1, corresponding tag;
Wherein x Sj Is the jth source domain sample data; y is Sj Is x Sj A corresponding label; j =1,2 s ;n s The number of the source domain samples is;
according to X S And X T Respectively constructing a kernel matrix K and a matrix L, and constructing a central matrix H;
wherein, K S,S ,K T,T ,K S,T ,K T,S Kernel functions of source domain data, target domain data and mixed domain data respectively;
wherein the content of the first and second substances,is a matrix of the unit, and is,is a full l column vector;
r is (n) of the full real domain s +n t )*(n s +n t ) An order matrix; l T A transposed matrix of l;
s202, matrix pair (KLK + mu I) -1 KHK carries out characteristic decomposition, and combines the eigenvectors corresponding to m maximum eigenvalues together to obtain a projection matrix W, wherein m<(k1,k2) min ,μ>0 is a trade-off parameter;
s203, enabling X to pass through W S And X T Projective transformation into a data set of a mapping spaceAndwhile maintaining the main characteristic attributes in the data sets of the source domain and the target domain, reducing the edge distribution difference between the domains by reducing the sample mean distance between the domains;
Wherein W T Is a transposed matrix of W.
4. The photovoltaic power prediction method based on feature migration according to claim 1, wherein the setting method of the fixed weight coefficient in step S6 is:
first, the absolute error of each single model is calculated, and the formula is as follows:
(ii) a Where Yi is the actual power at time i, P it The predicted power of the t-th single model at the moment i is obtained, and N is the total number of samples for solving the fixed weight coefficient;
5. A photovoltaic power prediction system based on feature migration, comprising:
the target value module is used for acquiring historical power generation data of the photovoltaic power station, removing abnormal values and taking the abnormal values as target values of model training;
the target domain module is used for acquiring historical prediction meteorological elements corresponding to historical power generation data of the photovoltaic power station from the numerical weather forecast, extracting characteristics related to time and geography from the meteorological elements and constructing target domain training sample data;
the source domain module is used for acquiring historical meteorological elements corresponding to historical power generation data of the photovoltaic power station from the ERA5 atmospheric re-analysis meteorological data, extracting characteristics related to time and geography from the meteorological elements and constructing source domain training sample data;
the TCA module is used for obtaining a training data set through source domain training sample data and target domain training sample data based on a migration component analysis method;
the training module is used for inputting the target value and the training data set into LSTM and Lasso models to be trained respectively, and training the models after gridding search and parameter adjustment to obtain the well-trained LSTM and Lasso models;
and the prediction module is used for respectively predicting the photovoltaic power through the trained LSTM model and the trained Lasso model, and performing photovoltaic power combined prediction on the prediction result by using a fixed weight coefficient to obtain a final prediction result.
6. The system according to claim 5, wherein the target domain module and the source domain module are both provided with a feature extraction unit, and the feature extraction unit comprises:
the extraction subunit is used for extracting meteorological elements;
the time characteristic subunit is used for extracting characteristics related to time, converting the timestamp information of the meteorological element data into discrete time information, and adding labels of months, weeks, days and hours to each meteorological element data;
and the geographic characteristic subunit is used for extracting geographic related characteristics and adding the inclination angle and longitude and latitude information of the photovoltaic square matrix, and the labels of the sine of the solar altitude angle and the cosine of the solar incident angle to each meteorological element data.
7. The feature migration based photovoltaic power prediction system of claim 5, wherein the TCA module comprises:
matrix unit: constructing a feature number of k 1 Target domain training sample data ofCorresponding label;
Wherein x is Ti Is the ith item of target domain sample data; y is Ti Is x Ti A corresponding label; i =1,2 t ,;n t The number of samples in the target domain is;
constructing a feature number of k 2 Source domain training sample data of,k2>k1, corresponding tag;
Wherein x Sj Is the jth source domain sample data; y is Sj Is x Sj A corresponding label; j =1,2 s ;n s The number of the source domain samples is;
according to X S And X T Respectively constructing a kernel matrix K and a matrix L, and constructing a central matrix H;
wherein, K S,S ,K T,T ,K S,T ,K T,S Kernel functions of source domain data, target domain data and mixed domain data respectively;
wherein the content of the first and second substances,is a matrix of the units,is a full l column vector;
r is all-real-field (n) s +n t )*(n s +n t ) An order matrix; l T A transposed matrix of l;
a characteristic decomposition unit: pair matrix (KLK + mu I) -1 KHK carries out characteristic decomposition, and combines the eigenvectors corresponding to m maximum eigenvalues together to obtain a projection matrix W, wherein m<(k1,k2) min ,μ>0 is a trade-off parameter;
a data set unit: by W making X S And X T Projective transformation into a data set of a mapping spaceAndwhile maintaining the main characteristic attributes in the data sets of the source domain and the target domain, reducing the edge distribution difference between the domains by reducing the sample mean distance between the domains;
Wherein W T Is a transposed matrix of W.
8. The feature migration based photovoltaic power prediction system of claim 5, wherein the prediction module comprises a fixed weight coefficient unit, the fixed weight coefficient unit comprising:
calculating the absolute error of each single model, wherein the formula is as follows:
(ii) a Where Yi is the actual power at time i, P it The predicted power of the t-th single model at the moment i is obtained, and N is the total number of samples for solving the fixed weight coefficient;
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210983674.2A CN115049171B (en) | 2022-08-17 | 2022-08-17 | Photovoltaic power prediction method and system based on feature migration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210983674.2A CN115049171B (en) | 2022-08-17 | 2022-08-17 | Photovoltaic power prediction method and system based on feature migration |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115049171A CN115049171A (en) | 2022-09-13 |
CN115049171B true CN115049171B (en) | 2022-10-28 |
Family
ID=83167263
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210983674.2A Active CN115049171B (en) | 2022-08-17 | 2022-08-17 | Photovoltaic power prediction method and system based on feature migration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115049171B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115347571B (en) * | 2022-10-17 | 2023-03-24 | 国网江西省电力有限公司电力科学研究院 | Photovoltaic power generation short-term prediction method and device based on transfer learning |
NL2033884B1 (en) * | 2022-12-30 | 2023-09-06 | Gansu Tongxing Intelligent Tech Development Co Ltd | A Photovoltaic Power Prediction Method, System and Medium |
CN116073436B (en) * | 2023-04-06 | 2023-06-13 | 山东创宇环保科技有限公司 | Capacity optimization control method for photovoltaic new energy power system |
CN116722548B (en) * | 2023-08-09 | 2023-12-29 | 深圳航天科创泛在电气有限公司 | Photovoltaic power generation prediction method based on time sequence model and related equipment |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110766212A (en) * | 2019-10-15 | 2020-02-07 | 哈尔滨工程大学 | Ultra-short-term photovoltaic power prediction method for historical data missing electric field |
WO2021238505A1 (en) * | 2020-05-27 | 2021-12-02 | 华北电力大学 | Federated learning-based regional photovoltaic power probability prediction method, and cooperative regulation and control system |
CN113919232A (en) * | 2021-10-25 | 2022-01-11 | 北京航天创智科技有限公司 | Photovoltaic power station power prediction method and system based on recurrent neural network |
CN114219181A (en) * | 2022-01-20 | 2022-03-22 | 天津大学 | Wind power probability prediction method based on transfer learning |
CN114462723A (en) * | 2022-04-12 | 2022-05-10 | 南方电网数字电网研究院有限公司 | Cloud layer migration minute-level photovoltaic power prediction method based on high-altitude wind resource influence |
CN114897264A (en) * | 2022-06-08 | 2022-08-12 | 国网浙江省电力有限公司嘉兴供电公司 | Photovoltaic output interval prediction method under small sample scene based on transfer learning |
-
2022
- 2022-08-17 CN CN202210983674.2A patent/CN115049171B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110766212A (en) * | 2019-10-15 | 2020-02-07 | 哈尔滨工程大学 | Ultra-short-term photovoltaic power prediction method for historical data missing electric field |
WO2021238505A1 (en) * | 2020-05-27 | 2021-12-02 | 华北电力大学 | Federated learning-based regional photovoltaic power probability prediction method, and cooperative regulation and control system |
CN113919232A (en) * | 2021-10-25 | 2022-01-11 | 北京航天创智科技有限公司 | Photovoltaic power station power prediction method and system based on recurrent neural network |
CN114219181A (en) * | 2022-01-20 | 2022-03-22 | 天津大学 | Wind power probability prediction method based on transfer learning |
CN114462723A (en) * | 2022-04-12 | 2022-05-10 | 南方电网数字电网研究院有限公司 | Cloud layer migration minute-level photovoltaic power prediction method based on high-altitude wind resource influence |
CN114897264A (en) * | 2022-06-08 | 2022-08-12 | 国网浙江省电力有限公司嘉兴供电公司 | Photovoltaic output interval prediction method under small sample scene based on transfer learning |
Non-Patent Citations (2)
Title |
---|
基于SDA与SVR混合模型的迁移学习预测算法;任俊 等;《计算机科学》;20180131;第45卷(第01期);第280-284、312页 * |
基于迁移成分分析的雷达辐射源识别方法研究;李蒙 等;《电子设计工程》;20171020;第25卷(第20期);第47-50、55页 * |
Also Published As
Publication number | Publication date |
---|---|
CN115049171A (en) | 2022-09-13 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN115049171B (en) | Photovoltaic power prediction method and system based on feature migration | |
Huang et al. | Hybrid deep neural model for hourly solar irradiance forecasting | |
Zhang et al. | A solar time based analog ensemble method for regional solar power forecasting | |
Yu et al. | An LSTM short-term solar irradiance forecasting under complicated weather conditions | |
Ceci et al. | Predictive modeling of PV energy production: How to set up the learning task for a better prediction? | |
JP6759966B2 (en) | How to operate the photovoltaic power generation system | |
Yesilbudak et al. | A review of data mining and solar power prediction | |
Melzi et al. | Hourly solar irradiance forecasting based on machine learning models | |
Alomari et al. | A predictive model for solar photovoltaic power using the Levenberg-Marquardt and Bayesian regularization algorithms and real-time weather data | |
Zhang et al. | Optimal graph structure based short-term solar PV power forecasting method considering surrounding spatio-temporal correlations | |
Yang et al. | An analog technique to improve storm wind speed prediction using a dual NWP model approach | |
Mas' ud | Comparison of three machine learning models for the prediction of hourly PV output power in Saudi Arabia | |
Kayri et al. | Predicting power production from a photovoltaic panel through artificial neural networks using atmospheric indicators | |
CN112070311A (en) | Day-ahead light power prediction method based on similar day clustering and meteorological factor weighting | |
Yildiz et al. | A kernel extreme learning machine-based neural network to forecast very short-term power output of an on-grid photovoltaic power plant | |
Wu et al. | Potential of Bayesian additive regression trees for predicting daily global and diffuse solar radiation in arid and humid areas | |
Wojtkiewicz et al. | A concept-drift based predictive-analytics framework: Application for real-time solar irradiance forecasting | |
Liu et al. | Research of photovoltaic power forecasting based on big data and mRMR feature reduction | |
Ahmad et al. | Support vector machine based prediction of photovoltaic module and power station parameters | |
Zhu et al. | [Retracted] Photovoltaic Generation Prediction of CCIPCA Combined with LSTM | |
CN113298304A (en) | Photovoltaic power generation power prediction method based on multi-dimensional weather forecast factor graph convolution | |
Bai et al. | Accurate four-hour-ahead probabilistic forecast of photovoltaic power generation based on multiple meteorological variables-aided intelligent optimization of numeric weather prediction data | |
Fu et al. | Photovoltaic Power Prediction Based on Gray-scale Satellite Cloud Image and Optimized LSTM | |
Pytel et al. | Use of Evolutionary Algorithm for Identifying Quantitative Impact of PM2. 5 and PM10 on PV Power Generation | |
Zhang et al. | Multi‐Input Deep Convolutional Neural Network Model for Short‐Term Power Prediction of Photovoltaics |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |